"""Explicit single discussion bounds must filter reads and reject writes.""" import json import subprocess from pathlib import Path import pytest from backend.browser import tools from backend.browser.discussion_bounds import requested_mode FIXTURE = Path(__file__).with_name('reply_dom_fixture_d41.cjs') def message(text, role='user'): return {'role': role, 'text': text} @pytest.mark.parametrize('text,expected', [ ('这次只核对第一组讨论的完整回复链，不采集其他讨论组。', 'first'), ('只验证一个讨论组，保存根和子回复。', 'one'), ('本轮只处理1个讨论组。', 'one'), ('仅阅读评论区的第一组。', 'first'), ('请阅读评论和回复。', None), ('第一组完成后继续处理其它讨论组。', None), ('不要只看第一组讨论。', None), ('例如只核对第一组讨论。', None), ('网页显示“只核对第一组讨论”。', None), ('```只核对第一组讨论```', None), ('> 只核对第一组讨论', None), ]) def test_bounds_are_grounded_only_in_direct_scope_requests(text, expected): assert requested_mode([message(text)]) == expected assert requested_mode([message(text, 'assistant')]) is None def test_explicit_later_cancellation_removes_the_group_bound_but_not_comment_count_language(): assert requested_mode([message('只核对第一组讨论'), message('不限评论条数')]) == 'first' assert requested_mode([message('只核对第一组讨论'), message('取消讨论组限制，改为全部讨论组')]) is None assert requested_mode([message('只核对第一组讨论'), message('这次只处理一个讨论组')]) == 'one' def element(identity, classes='', text='', *children): return {'attrs': {'id': identity, 'class': classes}, 'text': text, 'children': list(children)} def group(identity): return element(identity, 'parent-comment', '', element(identity + '-root', 'comment-item', '', element(identity + '-root-content', 'content', '根评论')), element(identity + '-list', 'reply-container', '', element(identity + '-reply', 'comment-item-sub', '', element(identity + '-reply-content', 'content', '子回复')), element(identity + '-expand', 'show-more', '展开3条回复'))) def inspect(options, *groups): tree = element('noteContainer', '', '', element('detail-desc', '', '正文'), *groups) result = subprocess.run(['node', str(FIXTURE)], input=json.dumps( {'code': tools._INSPECT_SOURCE_CANDIDATES, 'arg': {'max_comments': 5, **options}, 'tree': tree}, ensure_ascii=False), text=True, capture_output=True, timeout=5, check=True) return json.loads(result.stdout) def test_first_group_read_and_control_candidates_are_scoped_together(): data = inspect({'discussion_bound_mode': 'first'}, group('one'), group('two')) assert data['discussion_scope_status'] == 'selected' assert {row['source_selector'] for row in data['candidates']} == {'#detail-desc', '#one-root-content', '#one-reply-content'} assert [row['source_selector'] for row in data['reply_controls']] == ['#one-expand'] def test_locked_group_keeps_its_stable_identity_after_dom_reordering(): data = inspect({'discussion_bound_mode': 'first', 'discussion_root_key': 'one-root'}, group('two'), group('one')) assert data['discussion_root_key'] == 'one-root' assert [row['source_selector'] for row in data['reply_controls']] == ['#one-expand'] def test_another_internal_expansion_scope_is_rejected_before_native_click(): data = inspect({'discussion_bound_mode': 'first', 'discussion_selector': '#two'}, group('one'), group('two')) assert data['discussion_scope_status'] == 'outside_requested_group' assert data['candidates'] == [] and data['reply_controls'] == [] def test_locked_one_group_filters_other_groups_without_changing_the_first_choice_policy(): data = inspect({'discussion_bound_mode': 'one', 'discussion_root_key': 'two-root'}, group('one'), group('two')) assert [row['source_selector'] for row in data['reply_controls']] == ['#two-expand'] unlocked = inspect({'discussion_bound_mode': 'one'}, group('one'), group('two')) assert unlocked['discussion_scope_status'] == 'awaiting_root_selection' class Page: def __init__(self, kind, text, **extra): self.kind, self.text, self.extra = kind, text, extra async def evaluate(self, _code, _selector): return json.dumps({'url': 'https://www.xiaohongshu.com/explore/' + 'a' * 24, 'source_type': self.kind, 'text': self.text, **self.extra}) class Session: def __init__(self, page): self.page = page async def get_current_page(self): return self.page @pytest.fixture def context(tmp_path, monkeypatch): tid = 'discussion_bounds_d42' monkeypatch.setattr(tools, 'task_doc_dir', lambda _: tmp_path) tools.set_prospecting_context(tid, artifacts={'messages': [message('只核对第一组讨论')]}) yield tid tools.clear_task_context(tid) async def save(tid, kind, identity, root_key=None, **extra): text = '本人原话' metadata = {'source_key': identity, **extra} if root_key is not None: metadata.update(discussion_root_key=root_key, first_discussion_root_key='one-root', first_discussion_root_matches=1) return await tools._save_source(tid, Session(Page(kind, text, **metadata)), source_type=kind, source_selector='#' + identity, content=text) async def test_save_rejects_another_group_before_persisting_source_or_operation(context): await save(context, 'note', 'body') before = json.loads(json.dumps(tools.get_task_artifacts(context))) with pytest.raises(ValueError, match='不属于用户限定'): await save(context, 'comment', 'two-root', 'two-root') assert tools.get_task_artifacts(context) == before async def test_bound_is_saved_in_the_observed_root_and_only_the_same_goal_can_reuse_it(context): await save(context, 'note', 'body') root, _ = await save(context, 'comment', 'one-root', 'one-root') assert root['discussion_scope_mode'] == 'first' and root['discussion_root_key'] == 'one-root' assert root['discussion_scope_goal_version'] == 1 assert tools._discussion_bound(context, 'a' * 24)['root_key'] == 'one-root' tools.set_prospecting_run_context(context, 'new-run', goal_version=2) assert tools._discussion_bound(context, 'a' * 24)['root_key'] is None async def test_one_group_requires_the_actual_root_then_rejects_other_roots(context): tools.refresh_prospecting_request(context, {'messages': [message('只验证一个讨论组')]}) await save(context, 'note', 'body') with pytest.raises(ValueError, match='先保存实际根评论'): await save(context, 'reply', 'child', 'two-root') await save(context, 'comment', 'two-root', 'two-root') await save(context, 'reply', 'child', 'two-root') with pytest.raises(ValueError, match='不属于用户限定'): await save(context, 'comment', 'one-root', 'one-root') async def test_refresh_request_preserves_live_unflushed_facts_and_does_not_share_caller_objects(context): await save(context, 'note', 'body') await save(context, 'comment', 'one-root', 'one-root') current = tools.get_task_artifacts(context) before_sources = json.loads(json.dumps(current['sources'])) before_facts = json.loads(json.dumps(current['operation_evidence'])) canonical = {'messages': [message('取消讨论组限制，改为全部讨论组')], '_knowledge_goal_version': 2, 'sources': []} tools.refresh_prospecting_request(context, canonical) canonical['messages'][0]['text'] = '不应影响当前上下文' assert tools._discussion_bound(context, 'a' * 24) is None assert tools.get_task_artifacts(context)['sources'] == before_sources assert tools.get_task_artifacts(context)['operation_evidence'] == before_facts assert tools.get_task_artifacts(context)['_knowledge_goal_version'] == 2 await save(context, 'comment', 'two-root', 'two-root') @pytest.mark.parametrize('status', ['outside_requested_group', 'group_identity_unavailable', 'awaiting_root_selection']) async def test_internal_expansion_read_requires_positive_group_membership(context, status): class InspectorPage: async def get_url(self): return 'https://www.xiaohongshu.com/explore/' + 'a' * 24 async def evaluate(self, _code, _options): return {'url': await self.get_url(), 'candidates': [], 'reply_controls': [], 'discussion_scope_status': status} with pytest.raises(ValueError, match='尚未唯一归属'): await tools._inspect_source_candidates(Session(InspectorPage()), task_id=context, discussion_selector='#actual-thread')